Laser projector assembly

Abstract

The invention describes a laser projector assembly (1) for use in a wind turbine rotor blade manufacturing apparatus (3) comprising a plurality of laser projector units (10) wherein each laser projector unit (10) comprises positioning means (13) for positioning the laser projector unit (10) above a selected rotor blade mold (2), and a laser projector (12) configured to project a layup guide (12G) into that mould (2) during the manual layup procedure; an imaging arrangement (11) adapted to capture an image (110) of that mould (2); a machine learning algorithm (18) trained to determine coordinates of the features (2M, 12P) in the image (110); and a calibration module (16) configured to calibrate the laser projector (12) to that mould (2) prior to the manual layup procedure on the basis of the output (180) of the machine learning algorithm (18). The invention further describes a method of manufacturing a wind turbine rotor blade (4) using such a laser projector assembly (1), a machine learning algorithm (18) for use in such a laser projector assembly (1) and a method of training such a machine learning algorithm.

Description

Background Art

[0001] Wind turbine rotor blades can be manufactured by arranging sheets of composite material (usually fiberglass material) together with structural elements (such as spars and spar caps) into a mold and then closing the mold. Resin is injected into the closed mold using a system of inlets, outlets and vacuum pumps so that the fluid resin is evenly distributed throughout the composite material layers. The resin is then allowed to cure, a process that can be accelerated by heating elements embedded in the mold body. Finally, the mold is opened to release the cured rotor blade or rotor blade shell.

[0002] During a layup procedure, the composite material sheets can be arranged in the mold according to a predefined layup plan, which specifies which type of composite material, the size of the sheets and the position of the sheets in the mold. In the case of long rotor blades (e.g., having a length of 80 m or more), it is substantially not feasible to carry out the layup in a completely manual procedure, especially in manufacturing facilities that produce rotor blades in large quantities.

[0003] It is known to deploy laser projector assemblies to assist in the layup procedure. For example, a fixed laser projector assembly includes multiple laser projectors mounted above a mold in a manufacturing facility, and one laser projector may be required every 6-8 m. Each laser projector is assigned to a mold section and is used to project the outline of a composite sheet into the mold, showing the expected position of the composite sheet within that mold section. A technician following the layup plan for that mold can then place the composite sheet in the correct position.

[0004] This type of laser projector assembly must be carefully calibrated to that blade rotor mold in order to ensure that the projected profile corresponds precisely to the position of the composite material sheet. This requirement can in turn result in additional expense costs for a number of reasons. For example, the laser projector assembly can only be used with a mold for which it is specifically calibrated. A manufacturing facility that produces multiple series of rotor blades will require a dedicated projector assembly for each series, as the various sizes of the molds will be different. A fixed laser projector assembly for each of the multiple molds is very expensive and contributes significantly to the cost of rotor blade manufacturing.

[0005] Furthermore, because the molds currently used in the large-scale manufacture of wind turbine rotor blades are also substantially made of synthetic materials (such as fiberglass), the molds themselves can gradually shrink during their service life. The final or final length of a "used" mold can be significantly shorter than its initial length. Therefore, when using prior art laser projector assemblies, a labor-intensive manual calibration substantially precedes the layup procedure in order to adjust the laser projector as needed. These measurement and calibration steps can be time consuming and require high-precision measurement procedures in order to avoid layup errors. Known methods therefore contribute significantly to the overall cost of manufacturing rotor blades.

[0006] It is therefore an object of the present invention to provide an improved laser projector assembly which overcomes the problems outlined above.

[0007] This object is achieved by a claimed laser projector assembly; by a claimed method of manufacturing a wind turbine rotor blade; and by a claimed machine learning algorithm. Summary of the invention

[0008] The rotor blades of current wind turbines can have a length of about 80m-100m or even more. In the following, it may be assumed that a mold is used in the manufacture of such a long rotor blade. The laser projector is used to project a contour into the mold during the manual layup procedure; to show the position, size and shape of the composite sheets to be placed; and the projected contour may be referred to herein as a layup guide, contour guide, guide contour, etc.

[0009] The laser projector assembly described hereinafter should be understood as being intended for use in a manufacturing facility for a series production of wind turbine rotor blades. The manufacturing facility may comprise a building or plant large enough to accommodate a plurality of rotor blade moulds and other equipment required for carrying out at least the layup, vacuum-assisted resin transfer, curing and demoulding stages simultaneously. In the context of the present invention, the term "selected mould" can be understood as the "lower half" or "lower shell" of the mould, which is used in the closed mould manufacturing technique. The entire rotor blade layup is placed in this lower shell prior to closing the mould as explained in the introduction. Likewise, the "selected mould" can be one of two moulds used in the "butterfly" moulding technique, in which each rotor blade half is cast separately and the cured halves are joined along their outer periphery.

[0010] According to the present invention, the laser projector assembly includes a plurality of laser projector units and a positioning device for moving the laser projector unit into a position on a selected mold. The laser projector unit includes a laser projector configured to project a layup guide into the selected mold during a manual layup procedure. The laser projector assembly also includes an imaging arrangement having a plurality of cameras arranged to capture images of the selected mold. The laser projector assembly further includes a module having a machine learning algorithm that has been trained to determine the coordinates of features in the image; and a calibration module that is suitable for calibrating the laser projector unit to that selected mold prior to the manual layup procedure based on the output of the machine learning algorithm.

[0011] An advantage of the inventive laser projector assembly is that each laser projector unit can operate substantially independently of any other laser projector unit. A laser projector unit that is no longer needed for a layup in one mould can be moved to the next mould. Another advantage is that the laser projectors of the laser projector units can be calibrated quickly and with high accuracy in a very short time, for example in the space of only a few seconds, so that the lead time prior to the layup procedure can be kept to a very advantageous minimum. Furthermore, the calibration of the laser projectors does not require any manual input, such as measurements obtained by a person or technician, which has the advantage of saving labour costs. Because the laser projector assembly, as will be explained below, implements this calibration procedure substantially automatically, the procedure may be referred to herein as "automatic calibration".

[0012] According to the invention, a method for manufacturing a wind turbine rotor blade using an example of the inventive laser projector assembly comprises an initial step of moving the laser projector unit into a position above a selected mold (i.e., into a position for projecting guide profiles for the next stage of layup); operating an imaging arrangement device in order to capture an image of the selected mold; and applying a calibration unit in order to calibrate the laser projector unit to the selected mold in preparation for the layup stage. In a subsequent stage, the method comprises the step of operating the calibrated laser projector unit in order to project layup guides into the selected mold in order to assist the manual layup procedure.

[0013] According to the present invention, a trained machine learning algorithm for use in such a laser projector assembly comprises a machine learning module comprising an input layer, an output layer, and a plurality of intermediate layers. The input layer is configured to receive an image of a mold, which includes a target identification and a laser calibration pattern as will be explained below. The output layer is configured to output the coordinates of the target identification and the calibration pattern. This information is then used to make any necessary adjustments to the layup plan so that the laser projector will project the correct guide profile into that mold.

[0014] The terms "machine learning", "artificial intelligence" and "AI" should be understood to be synonymous and may be used interchangeably herein. In a preferred embodiment of the present invention, the machine learning module can be a neural network module. Any suitable neural network can be trained and deployed in the inventive laser projector assembly. In the following, without limiting the present invention in any way, it can be assumed that the machine learning includes a neural network, such as a convolutional neural network.

[0015] The laser projector is capable of projecting a beam of light precisely at the desired point. As will be known to those skilled in the art, the laser projector can be used to project a 2D image by providing image parameters to it and controlling the servo system of the scanner so that a mirror mounted at the end of a rotating shaft guides the beam in order to "draw" the image. Because of the high scanning speed, the viewer perceives the complete image, in this case, the outline of a sheet of synthetic material. For example, the image parameter can be the size of a rectangular pad or sheet of fiberglass material, and the laser projector is used to project the corresponding rectangular outline into the mold. When working on the "bottom half" of the layup for closed mold technology, the outline is projected into the concave lower shell. After the placement of the web, spar cap and mandrel, work continues on the "upper half" of the layup and the outline is projected onto the upper layup. After completion of the layup, the mold is closed and resin injection can be implemented.

[0016] Particularly advantageous embodiments and features of the invention are given by the dependent claims, as disclosed in the following description. Features of different claim categories may be combined as appropriate to give further embodiments not described herein.

[0017] The imaging arrangement of the laser projector unit can be configured so that the camera captures a single image showing the entire mold. Alternatively, multiple images (each showing a portion of the mold) can be captured, and the imaging arrangement can include an image stitching module that produces a composite image of the entire mold.

[0018] As explained above, the length of fiberglass rotor blade molds shrinks during their service life. In the invented method, the AI ​​algorithm enables the layup plan to be adjusted very accurately to correspond to the actual mold dimensions, thereby correcting to allow for any spanwise length difference between the assumed mold length and the actual mold length. As used in the context of the present invention, the phrase "calibrating the laser projector to the mold" should be understood in the generally accepted sense to mean adjusting the layup plan to correct for the difference between the assumed mold length and the actual mold length.

[0019] In a particularly preferred embodiment of the invention, a plurality of target markings are arranged on the mould so that these are visible to the imaging arrangement. The target markings should be understood as being clearly visible to the image sensor. For example, the target markings can have a clearly visible shape and / or can display a clearly discernible pattern. The target markings are preferably arranged at the upper face of the mould, for example on a flange around the periphery of the mould. The term "upper face" should be understood as an upwardly facing horizontal surface extending around the periphery of the mould.

[0020] The target markers are used during the training procedure and also during the subsequent calibration procedure and should be clearly discernible to the image processing algorithm. In an advantageously simple method, the target markers can be stickers with colors and / or patterns placed on the upper face of the mold in a predetermined arrangement, which clearly distinguish them from the mold optically. In this method, the target markers can be placed manually, for example at predefined leaf span intervals along the upper face. In a preferred embodiment of the invention, bushings that already exist around the periphery of the lower shell can be used as target markers. These bushings receive fasteners that are used to couple the upper shell to the lower shell prior to the closed mold resin injection phase. In order to make the bushings clearly discernible as target markers (for the image processing algorithm), in a preferred embodiment of the invention, the inner surface of the bushings can be painted with a suitable material (such as a reflective white paint).

[0021] Although the target markers are provided on the mold, they can be considered as part of the laser projector assembly because the sole purpose of the target markers is to assist the machine learning algorithm in calibrating the laser projector unit into the selected mold. The image of the mold can include one or more clearly identifiable target markers along the perimeter, such as markers placed at 10m intervals on a new mold, and the trained machine learning algorithm can infer the actual coordinates of the markers in the reference frame of that mold. As explained above, the length of the fiberglass mold can shrink noticeably during its service life. In the laser projector assembly of the invention, the machine learning algorithm is configured to determine the actual coordinates of any target marker in the image. In this way, the controller of the laser projector unit can establish its span position relative to the mold. For example, markers initially placed at 10m intervals in the span direction on a new mold may eventually be separated by 9.95m when the mold has reached its final length. This difference can be detected by controlling the laser projector to project a laser calibration pattern at the assumed coordinates of the target marker and by detecting any span offset between the target marker and the calibration pattern in the image.

[0022] In particular, in the case of a series production of wind turbine rotor blades, the layup procedure can be simplified by compiling a layup plan for a specific mold type. Such a layup plan can specify the order of placement of multiple synthetic material sheets between the large annular root end and the outermost tip end. The layup plan can specify the size and shape of each synthetic material sheet. The layup plan can also specify the type of each synthetic material sheet, such as CSM (cut chopped strand mat), slub yarn, unidirectional fabric, etc. The layup plan can also specify the expected position of structural parts (such as spar caps, shear webs, etc.) so that corresponding guide contours can be projected into the mold to assist the layup team in placing that part. Most relevant to the invention, the layup plan also specifies the position of each synthetic material sheet in the mold, and each such position is specified on the basis of an assumed and possibly incorrect mold length. This is because the layup plan is compiled under the assumption that the mold has a specific length, which is referred to herein as the "assumed mold length."

[0023] In a particularly preferred embodiment of the invention, the laser projector assembly is configured to receive such a layup plan customized for a selected mold, and the inventive machine learning algorithm is trained to determine any necessary calibrations that may need to be made to the layup plan so that the layup matches the selected mold.

[0024] The rotor blade mold is prepared by placing target markers at specific locations along the perimeter of the mold, or by painting an already existing bushing with a reflective paint as described above. A laser projector unit is placed above the mold, the laser projector is actuated to project a laser calibration pattern onto the mold, and an imaging arrangement is actuated to capture a set of images of the mold (a single image of the entire mold or a composite image obtained by image stitching). The set of images for that mold is referred to herein as an image batch. The training procedure for the machine learning algorithm includes the steps of obtaining a batch of images for the rotor blade mold and annotating the target markers and calibration patterns visible in the images. The images are also annotated with the coordinates of the target markers and calibration patterns (in a predefined reference system for the mold). The actual length of the mold is measured and represented as a ground truth for the machine learning algorithm. With these as input ground truths, the neural network learns to recognize the target markers and calibration patterns in the image, and learns to derive their coordinates. This supervised learning phase is repeated for multiple image batches until the accuracy of the neural network is satisfactory. The trained neural network can then receive an image batch as input and will output the coordinates of any visible target markers and calibration patterns. The training dataset for the machine learning algorithm can include batches of several thousand annotated images for any one type of rotor blade mold, depending on the required / desired accuracy of the calibration (the accuracy can be improved by using more images to train the algorithm).

[0025] Once the laser projector unit has been moved into position on that mold, the trained neural network can then be used to calibrate the laser projector unit to the selected mold. In a first step, one or more cameras in the laser projector assembly are operated to capture an image of the mold while the laser projector is controlled to project a calibration pattern at a specified point (e.g., onto the desired location of a target mark). The neural network processes the image and outputs the coordinates of that target mark and the laser calibration pattern. Any difference between the expected laser position and its actual position (e.g., it may be "off" by a few millimeters or even centimeters in the spanwise direction) is then used to calibrate the laser projector unit to that mold by correcting the entries in the layup plan by corresponding factors.

[0026] The calibration procedure takes into account any differences between the "actual" mold length and the assumed mold length (the layup plan is based on the assumed mold length). For example, as explained above, the actual length of a new or relatively new mold may be slightly longer than the assumed mold length. If the layup plan is based on an incorrectly assumed mold length, following this layup plan without any corrective adjustments will have the result that the completed layup will not reach the tip in that selected mold. In this case, the calibration of the laser projector assembly to that mold with the correction of the layup plan will ensure that each contour guide is "scaled up", i.e., corrected to adjust for the "excess" mold length. In another example, the actual length of an older mold may be significantly shorter than the assumed mold length. If the layup plan is followed without any corrective adjustments, the completed layup will not reach the tip of the selected mold. In this case, the calibration of the laser projector assembly to that mold with the correction of the layup plan will ensure that the length of each contour guide is "scaled up", i.e., corrected to adjust for the "difference" in mold length.

[0027] The laser projector assembly can be implemented in any suitable manner in the manufacturing facility. For example, any camera of the imaging arrangement can be wall mounted, track mounted, or suspended from a cable arrangement. The camera can be fixed (i.e., its position remains unchanged) or movable so that the camera can be moved into position as needed.

[0028] Likewise, the laser projector unit can be wall mounted, track mounted, etc. In a particularly preferred embodiment of the invention, the laser projector unit is configured to move along a track mounted to the ceiling of the manufacturing facility. A row of rotor blade molds can be arranged below the track, for example in a configuration in which the major axis of the bladed rotor is aligned with the track. The positioning device of each laser projector unit preferably includes a suitable motor driven actuator, which is suitable for moving the laser projector unit along the track. The actuator can be controlled in an overall automatic manner. For example, following the layup plan, the positioning device can move the laser projector unit further in the direction of the tip end so that the next section of the layup can be implemented.

[0029] In the case of long rotor blades, multiple workers can perform manual layup, and laser projector units can be deployed to assist teams of one or more workers. For example, on a relatively short or narrow section of the mold, one laser projector can be deployed to assist the layup team working on that section. For relatively long or wide sections, more than one laser projector can be deployed to assist the layup team. As the layup approaches completion, any "free" laser projector unit can move or be moved to another planned position on the mold. In this way, multiple laser projector units can be optimally utilized while keeping the total number of laser projector units to a beneficial minimum. The total number of laser projector units can depend on the number of workers or teams, since the purpose of the laser projector units is to assist workers in performing manual layup. In a preferred embodiment of the method of the invention, the number of laser projector units deployed at any one time corresponds to the number of teams working on the layup procedure. For example, a single laser projector unit can be deployed to assist a single team working on the layup procedure; three laser projector units can be deployed when three layup teams are working on different sections of the mold at the same time.

[0030] The inventive laser projector assembly advantageously allows for seamless "handoff" from one laser projector unit to another. For example, if three laser projector units have been used to assist the layup team, and the middle unit is no longer needed, one of the outer laser projector units can be "handed off" to the middle laser projector unit and then moved to the next mold, where the calibration procedure as described above is ready to be performed for the layup in that next mold.

[0031] As explained in the introduction, prior art laser projector installations may require a fixed arrangement of laser projectors for specific sections of the lower mould, each mounted so as to project a contour guide into a corresponding lower mould section. The projector is calibrated for that specific lower mould until it has reached the end of its useful life. The fixed laser projector installation remains idle during the resin injection and curing steps, and cannot be used until the mould is opened and the cured rotor blade is removed from the lower shell. In contrast, the laser projector units of the invented laser projector assembly are capable of substantially continuous use because each laser projector unit can be moved to a new position on any rotor blade mould below the track. Because the laser projector units are independently movable, fewer units are required for the task of projecting contour guidance into a selected mould. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Other objects and features of the present invention will become apparent from the following detailed description considered in conjunction with the accompanying drawings. It should be understood, however, that the drawings are designed for illustration purposes only and are not intended as a definition of the limits of the invention.

[0033] Figure 1 Figure 1 illustrates a plurality of molds arranged in a wind turbine rotor blade manufacturing facility;

[0034] Figure 2 and 3 Demonstrate calibration procedures;

[0035] Figure 4 Demonstrating post-casting shrinkage of rotor blade molds;

[0036] Figure 5 Demonstrating post-cure shrinkage of a cured rotor blade;

[0037] Figure 6 Demonstrate training procedures for invented machine learning algorithms;

[0038] Figure 7 Demonstrate the stages in a training or calibration procedure;

[0039] Figure 8 A prior art rotor blade manufacturing apparatus is shown.

[0040] In the drawings, like reference numerals refer to like objects throughout. Objects in the drawings are not necessarily drawn to scale. DETAILED DESCRIPTION

[0041] Figures 1 to 3 Example aspects of the inventive laser projector assembly 1 are shown. Figure 1 The figure shows a plurality of molds 2 arranged in a wind turbine rotor blade manufacturing apparatus 3, while Figure 2 and Figure 3The laser projector unit 10 is shown in a selected position on the mold 2 .

[0042] exist Figure 1 , manual lay-up in different stages of completion in some of the moulds 2 is shown (by the shaded areas) while the other moulds 2 remain empty. The laser projector assembly 1 comprises a plurality of laser projector units 10 movably mounted on a track 13. The track 13 can, for example, be mounted on or suspended from the ceiling of a manufacturing plant. Here, two tracks 13 are provided, each carrying a plurality of laser projector units 10, and a row of moulds 2 is arranged below each track 13. Each laser projector unit 10 comprises a plurality of laser projector units 10, each carrying ... Figure 2 1 and 14. Suitable positioning means for moving the laser projector unit 10 into position can be actuated, for example, by the controller. Figure 2 Also indicated is a camera of an imaging arrangement 11, which may be ceiling mounted, suspended from a cable arrangement or otherwise positioned above the mold 2. Here, the camera 11 is actuated so as to capture an image 110 of the mold 2 as indicated by the dashed line. The image (which can be a single image or a stitched composite image) also shows an arrangement of target markers 2M provided around the perimeter of the mold 2. In this exemplary embodiment, the target markers 2M are formed as a reflective coating in a bushing 20, which will later be used to connect the upper shell to the lower shell 2.

[0043] The laser projector unit 10 is also configured to receive the ply plan 2 铺层 , which specifies the order of placement of the composite material sheets in that mould 2. As explained above, the laser projector 12 is used during the subsequent manual layup procedure to project contour guides 12G (an example guide 12G is illustrated) for each composite material sheet into the lower mould 2 according to the layup plan, which can be provided as a suitable data format that will be known to those skilled in the art.

[0044] The laser projector assembly 1 in this example embodiment comprises a trained neural network 18. This receives an image 110 from one or more cameras (only one is illustrated here for simplicity) of an imaging arrangement 11, identifies any target identification 2M and calibration pattern 12P in the image and determines the coordinates for any such features 2M, 12P in a reference frame 2F for the mould. In an initial calibration step, the laser projector 12 is controlled to project the calibration pattern 12P at a specific location on the mould 2 (e.g. onto a specified target identification 2M), and simultaneously the imaging arrangement 11 is controlled to capture an image 110 showing the calibration pattern 12P. The image 110 (which will include a section of the mould 2, some target identification 200 and any calibration pattern 12P) is fed into the neural network 18. The neural network 18 can have been trained to output the coordinates of any target identification 2M and calibration pattern 12P seen in the image in the reference frame 2F for that mould 2. The controller 14 can then determine any spanwise differences ΔMP between the projected coordinates of the calibration pattern and the actual coordinates of the pattern 12P as the pattern 12P appears on the mold 2. If the actual length L2 of the mold 2 does not match the assumed mold length, the calibration pattern 12P will be offset from its expected position in the spanwise direction. The detection of the difference ΔMP may rely on previously determined tolerances, for example if the calibration pattern 12P is offset from the target mark 2M by less than 5 mm, the calibration pattern 12P and the target mark 2M may be considered to be consistent.

[0045] To conclude the automatic calibration step, the calibration module 16 adjusts the ply plan 2 by appropriate factors based on the output 180 of the neural network 18. 铺层 For example, the detected difference ΔMP may indicate that the spanwise length of each composite sheet should be shortened by 0.1%, and the calibration module responds by adjusting each entry of the layup plan accordingly. Similarly, the spanwise position of any structural element (such as a spar cap) can also be corrected by an appropriate amount. Adjusted layup data 16 铺层 It is then used to control the laser projector 12, which continues to project highly accurate layup guidance to assist the layup team.

[0046] Once the laser projector unit 10 has completed the calibration sequence, the manual layup procedure can be performed as follows: Figure 3 The process starts as indicated in FIG. 1 and the laser projector 12 is used to “draw” a contour guide 12G in the mold 2 according to the ply plan. The ply scheduler or controller 14 is able to execute the (adjusted) ply plan 16 by controlling the laser projector 12. 铺层 , so that for each projection profile guide 12G of the sequence of synthetic material sheets, which are to be placed into that section of the mould 2, a control signal 120 is provided, for example, as illustrated here in the corrected ply data 16 铺层is published to the laser projector 12 on the basis of.

[0047] Figure 4 Mold 2 is shown, indicating its initial length L2 新 and its final length L2 最终 and the difference Δ2 between these "new" and "final" lengths. For example, a newly manufactured fiberglass mold 2 for a 100 m rotor blade may have a length of 100.5 m, the additional 50 cm being intended to account for the shrinkage of the rotor blade during curing. After being used to mold and cure many rotor blades, the mold length L2 最终 It can be reduced to 100.3m. Until the mold has completed its shrinkage process, the actual mold length can be within the initial length L2 新 and the final length L2 最终 Anywhere in between.

[0048] Figure 5 A simplified partial cross section through a cured rotor blade 4 is shown. The length L4 of the rotor blade 4 is slightly less than the actual length L2 of the mold 2, which is indicated by the difference ΔL4. However, the layup plan for that mold 2 may be based on the length L4 of the cured rotor blade 4 or on incorrect assumptions about the mold length. The calibration performed by the inventive laser projector unit 10 immediately before each layup procedure aligns the layup plan 2 according to the actual mold length L2. 铺层 “Proportional changes” are made to avoid layup errors that would otherwise be caused by incorrect mold length assumptions.

[0049] Figure 6 is a flowchart showing the training procedure of the neural network. Image batch 110 for the mold 集 The batch 110 is obtained and input to the neural network in stage 61. As explained above, 集 The image will show a section of the mold, which includes the mold perimeter, the target mark 2M and the calibration pattern 12P. Each batch 110 集 The images of the batch 110 are annotated to identify the target marker 2M and the correction pattern 12P, and to include the coordinates of each target marker 2M and the correction pattern 12P. 集 , further related reference truth 610, such as the end-to-end length of the corresponding mold 2, is also input into the neural network

[0050] In stage 62, the neural network learns to distinguish batch 110 集The target mark 2M and the calibration pattern 12P in the image are obtained by performing the neural network learning. These stages 61, 62 are repeated a number of times until the neural network can reliably distinguish the target mark 2M and the calibration pattern 12P in the image, regardless of whether they are independently visible or whether the calibration pattern is superimposed on the target mark. The decision to continue ("no") or end ("yes") this learning stage is made in step 63.

[0051] When the neural network has learned to recognize the target mark and the calibration pattern with high accuracy, the training can continue to the next stage 64, in which the neural network learns to determine the coordinates of the target mark 2M or the calibration pattern 12P in the reference frame of the mold. The coordinates of the target mark 2M on the perimeter of the mold are known as a reference fact, and the neural network learns to deduce the coordinates of any target mark 2M visible in the image. Similarly, the neural network learns to deduce the coordinates of any calibration pattern 12P visible in the image. The decision to continue ("no") or end ("yes") this learning stage is made in step 65.

[0052] When the neural network has learned to deduce the coordinates of the target identification and calibration pattern with high accuracy, training can end at step 66. The neural network can now be used by the laser projector unit to determine its position relative to the mold, and to calibrate that laser projector unit to that mold.

[0053] Figure 7 A stage in a training or calibration procedure is shown and illustrates how a target mark 2M and a laser calibration pattern 12P may appear in an image of a rotor blade mould 2. Here, the target mark 2M has a distinctive pattern. If the mould dimensions remain unchanged over its service life, the laser calibration pattern 12P will be aligned with the mark 2M. However, due to inevitable blade span shrinkage, the laser calibration pattern 12P may shift from the mark 2M to some extent. Any such shift ΔMP is detected during calibration and the layup plan can be adjusted accordingly as explained above.

[0054] Figure 8 The problem known from the prior art is presented. The diagram illustrates a fixed laser assembly 8 in position on a rotor blade mould 2. The laser assembly 8 comprises a plurality of laser projectors 82. Each laser projector 82 is assigned to a section of the mould 2 and projects a profile 8G of the composite sheet into the mould 2, which has the purpose of indicating the correct position to the technician. However, since the initial length L2 新Due to the post-casting mold shrinkage to the actual length L2 (with the resulting difference Δ2), the projected profile will not correspond exactly to the optimal position of that composite sheet. Without correction, layup errors will accumulate. To avoid this problem, the prior art method requires careful manual measurement of the mold prior to each layup stage, as well as meticulous recalibration of the laser projector. The conflicting requirements of the necessary "proportional reduction" to take into account the thermal shrinkage of the mold and the necessary "proportional enlargement" to take into account the shrinkage of the composite plies during curing can make it very time-consuming to correctly install the laser projector prior to the manual layup procedure, increasing the cost of manufacturing. The initial calibration procedure immediately following the installation of a new fixed laser assembly as known from the prior art can take up to six hours, and each subsequent calibration prior to molding a new rotor blade can take up to one hour. In addition, the known laser projector assembly requires many laser projectors, for example one for every 7m of mold length. However, as illustrated here, only some are needed at any one time, while others are no longer needed (the laser projector on the left) or are not yet needed (the three on the right), i.e., the expensive equipment is not optimally utilized.

[0055] Although the present invention has been disclosed in terms of preferred embodiments and variations thereon, it will be appreciated that numerous additional modifications and variations can be made thereto without departing from the scope of the invention. For example, the inventive laser projector assembly and method of manufacture can be used in the manufacture of any large composite component that benefits from laser guidance during the layup procedure. Such a component can be a large structural element, such as a spar cap, for embedding in a rotor blade. Of course, the inventive laser projector assembly and method of manufacture can be adapted for use in the manufacture of large composite objects other than wind turbine rotor blades.

[0056] For reasons of clarity, it is to be understood that the use of "a" or "an" throughout this application does not exclude a plurality, and "comprising" does not exclude other steps or elements. The mention of a "unit" or "module" does not exclude the use of more than one unit or module.

Claims

1. A laser projector assembly (1) for use in a wind turbine rotor blade manufacturing apparatus (3), the laser projector assembly (1) comprising - a plurality of laser projector units (10), wherein each laser projector unit (10) comprises positioning means (13) for positioning the laser projector unit (10) above a selected rotor blade mould (2), and a laser projector (12) configured to project a layup guide (12G) into the mould (2) during a manual layup procedure; - an imaging arrangement (11) adapted to capture an image (110) of the mould (2); - a machine learning algorithm (18) trained to determine the coordinates of features (2M, 12P) in the image (110); and - a calibration module (16) configured to calibrate the laser projector (12) to the mold (2) based on the output (180) of the machine learning algorithm (18) prior to the manual layup procedure.

2. A laser projector assembly according to the preceding claim, wherein the machine learning algorithm (18) is a convolutional neural network.

3. A laser projector assembly according to any one of the preceding claims, wherein the output (180) of the machine learning algorithm (18) is a leaf span distance (ΔMP) between a target marker (2M) and a laser calibration pattern (12P).

4. The laser projector assembly according to any of the preceding claims, comprising a plurality of target markings (2M) provided at predetermined coordinates around the periphery of the rotor blade mould (2).

5. The laser projector assembly according to any one of the preceding claims, wherein the target marking (2M) comprises a reflective coating on the inner surface of the mold bushing (20).

6. The laser projector assembly according to any one of the preceding claims, configured to receive a layup plan (2) for a selected mold (2). 铺层 ).

7. The laser projector assembly according to any one of the preceding claims, wherein the layer plan (2 铺层 ) determine the order of placement of multiple composite sheets in the mold and / or the shape of each composite sheet and / or the type of each composite sheet and / or the position of each composite sheet in the mold.

8. The laser projector assembly according to claim 6 or claim 7, wherein the calibration module (16) is configured to adjust the ply plan (2M) based on a spanwise offset (ΔMP) between the target mark (2M) and the calibration pattern (12P). 铺层 ) entries.

9. The laser projector assembly according to any one of the preceding claims, wherein the imaging arrangement (11) comprises a plurality of cameras arranged above the mould (2).

10. Method for manufacturing a wind turbine rotor blade (4) using a laser projector assembly (1) according to any one of claims 1 to 9, comprising the following steps: - moving the laser projector unit (10) into position above the selected mould (2); and prior to the manual lay-up procedure, - operating the imaging arrangement (11) of the laser projector assembly (1) so as to capture an image (110) of the selected mould (2); and - applying the machine learning algorithm (18) to the image (110); - calibrating the laser projector unit (10) to the mould (2) based on the machine learning algorithm output (180); and then - operating the laser projector (12) of the calibrated laser projector unit (10) to project a layup guide (12G) into the selected mould (2).

11. The method according to the preceding claim, wherein the machine learning algorithm (18) determines the coordinates of the target identification (2M) and the laser calibration pattern (12P) shown in the captured image (110).

12. The method according to claim 10 or claim 11, comprising receiving a ply plan (2) for the selected mold (2). 铺层 ), and wherein the step of calibrating the laser projector unit (10) comprises adjusting the ply plan (20) based on the output (180) of the machine learning algorithm (18). 铺层 ).

13. A machine learning algorithm (18) for use in a laser projector assembly (1) according to any one of claims 1 to 9, comprising: - a neural network comprising an input layer, an output layer and a plurality of intermediate layers, wherein The input layer is configured to receive an annotated image (110) of the mold (2) 标签 ), the target mark (2M) and the laser calibration pattern (12P) have been labeled in the annotated image; and - the output layer is configured to provide the coordinates of the target mark (2M) and the laser calibration pattern (2P) in the reference system of the mould (2).

14. A method for training a machine learning algorithm according to claim 13, comprising the following steps: S0) arranging a laser projector unit (10) above the rotor blade mould (2); S1) obtaining an image set (110) of the mold (2); S2) annotating the image (110) by labeling the target marks (2M) and the laser calibration pattern (12P) and specifying the coordinates of those target marks (2M) and the laser calibration pattern (12P) in the reference system of the mold (2); S3) Repeating steps S1 and S2 in a supervised learning procedure in order to derive the coordinates of the target mark (2M) and the laser calibration pattern (12P) from the image (110) of the mold (2).

15. The method of claim 14, wherein the training data set comprises at least 2000 rotor blade mold image sets (110 集 ).

Images